![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F21739094%2F6fd0e169eb6cfc9d54aafcc921c9379b%2FScreenshot%202026-04-01%20010736.png?generation=1774987694210055&alt=media\")
Am I missing something? We selected this submission, but on the leaderboard same notebook is showing with 2 different scores a 1 is lower score and 1 is higher. Like were there multiple reruns of the same notebook After rerun of team’s two selected notebooks on a completely hidden test set, the final Private Leaderboard has posted!
\nThe leaderboard
\nWe are pleased to announce that team_cp @naganohikaru @yutaroito, AyPy @AyPy, and Stefan Stefanov @stefanstefanov are our three Leaderboard Prize winners. It was a close finish — congratulations to all who submitted!
\nWe were impressed that this competition elicited particularly strong publicly shared notebooks that many folks could build on and take advantage of.
\nSo it has been particularly exciting to now see several teams who were lower in the Public Leaderboard come out near the top of the Private Leaderboard — we are very curious to hear their strategies for generalization.
\nCompetition goals: Complex targets and beating the best_template_oracle
There were several goals of this Part 2 competition.
\nFirst, we wanted to test if automated modeling methods could generalize to RNAs without previously available templates and to other complex RNA targets, both aspects missing from Part 1.
\nBased on the results, the answer appears to be a resounding yes! We look forward to presenting detailed 3D structural comparisons later in 2026, when the ground truth structures will be made public.
\nSecond, we were curious if Kagglers could advance RNA structure prediction to, for the first time, consistently outperform a very hard baseline: the best_template_oracle.
This baseline was assembled from the best publicly available prior 3D template for each target – with special knowledge of the ground truth structure. In the Part 1 competition, Kaggle notebooks could not beat the best template oracle, even when all Kaggle structure predictions were pooled (see Part 1 preprint).
\nHere are the results from Part 2:
\nteam_cp + AyPy)best_template_oracleteam_cpAyPyThe top two scores from individual team_cp and AyPy notebooks are very close to the best template oracle — they are almost within statistical fluctuations (about 0.006).
Now look at the Top 2 model — it simply pools the predictions of one notebook each from team_cp and AyPy. Excitingly, this Top 2 outperforms the best template oracle! 🎉
Even better, if we take the Top 20 submitted notebooks, their collective best prediction for each target very convincingly outperforms the template oracle. Different top groups did better on different targets, demonstrating yet again the power of the Kaggle community to find diverse, complementary solutions to hard scientific problems.
\nWhat’s next?
\nIn just a couple weeks, the world’s flagship competition for molecular modeling will begin: the 17th Critical Assessment for Structure Prediction (CASP17).
\nCould this year mark a special jump in the accuracy of automated RNA structure prediction? The rapid advances in the last two Kaggle challenges certainly give us optimism — and your efforts in the next two weeks could make the difference:
\nThis is the time to make your notebooks public and to write up your solutions!
\nWe’re particularly interested in teams who placed in the Top 20 because, as noted above, the solutions appear complementary.
\nIf we share our insights and codes, we should be able to make an RNA structure prediction model that is better than the sum of its parts and increase our impact across biology, medicine, and biotechnology.
\nWe’re excited to again have collaboration across Kagglers, Stanford, HHMI, and NVIDIA for this ‘model synthesis’ effort and to enter resulting models into CASP17 for further rigorous blind tests. Looking forward to seeing you on the forums, in CASP, and in the RNA world!
\nFor the hosts,
\nRhiju Das @rhijudas and Przemek Porebski @przemekporebski
\nAnd, again, many thanks to our co-hosts and collaborators:
\nAfter rerun of team’s two selected notebooks on a completely hidden test set, the final Private Leaderboard has posted!
\nThe leaderboard
\nWe are pleased to announce that team_cp @naganohikaru @yutaroito, AyPy @AyPy, and Stefan Stefanov @stefanstefanov are our three Leaderboard Prize winners. It was a close finish — congratulations to all who submitted!
\nWe were impressed that this competition elicited particularly strong publicly shared notebooks that many folks could build on and take advantage of.
\nSo it has been particularly exciting to now see several teams who were lower in the Public Leaderboard come out near the top of the Private Leaderboard — we are very curious to hear their strategies for generalization.
\nCompetition goals: Complex targets and beating the best_template_oracle
There were several goals of this Part 2 competition.
\nFirst, we wanted to test if automated modeling methods could generalize to RNAs without previously available templates and to other complex RNA targets, both aspects missing from Part 1.
\nBased on the results, the answer appears to be a resounding yes! We look forward to presenting detailed 3D structural comparisons later in 2026, when the ground truth structures will be made public.
\nSecond, we were curious if Kagglers could advance RNA structure prediction to, for the first time, consistently outperform a very hard baseline: the best_template_oracle.
This baseline was assembled from the best publicly available prior 3D template for each target – with special knowledge of the ground truth structure. In the Part 1 competition, Kaggle notebooks could not beat the best template oracle, even when all Kaggle structure predictions were pooled (see Part 1 preprint).
\nHere are the results from Part 2:
\nteam_cp + AyPy)best_template_oracleteam_cpAyPyThe top two scores from individual team_cp and AyPy notebooks are very close to the best template oracle — they are almost within statistical fluctuations (about 0.006).
Now look at the Top 2 model — it simply pools the predictions of one notebook each from team_cp and AyPy. Excitingly, this Top 2 outperforms the best template oracle! 🎉
Even better, if we take the Top 20 submitted notebooks, their collective best prediction for each target very convincingly outperforms the template oracle. Different top groups did better on different targets, demonstrating yet again the power of the Kaggle community to find diverse, complementary solutions to hard scientific problems.
\nWhat’s next?
\nIn just a couple weeks, the world’s flagship competition for molecular modeling will begin: the 17th Critical Assessment for Structure Prediction (CASP17).
\nCould this year mark a special jump in the accuracy of automated RNA structure prediction? The rapid advances in the last two Kaggle challenges certainly give us optimism — and your efforts in the next two weeks could make the difference:
\nThis is the time to make your notebooks public and to write up your solutions!
\nWe’re particularly interested in teams who placed in the Top 20 because, as noted above, the solutions appear complementary.
\nIf we share our insights and codes, we should be able to make an RNA structure prediction model that is better than the sum of its parts and increase our impact across biology, medicine, and biotechnology.
\nWe’re excited to again have collaboration across Kagglers, Stanford, HHMI, and NVIDIA for this ‘model synthesis’ effort and to enter resulting models into CASP17 for further rigorous blind tests. Looking forward to seeing you on the forums, in CASP, and in the RNA world!
\nFor the hosts,
\nRhiju Das @rhijudas and Przemek Porebski @przemekporebski
\nAnd, again, many thanks to our co-hosts and collaborators:
\nI understand that auto-selection and final Private Leaderboard evaluation are supported by the rules. However, I do not think the rules made it sufficiently explicit that notebook submissions would be rerun for final evaluation, and that these reruns could lead to outcomes different from what participants inferred from the visible public notebook scores. That part seems to rely more on Kaggle’s notebook competition mechanics and forum clarifications than on a clearly stated rule.
", "rawMarkdown": "I understand that auto-selection and final Private Leaderboard evaluation are supported by the rules. However, I do not think the rules made it sufficiently explicit that notebook submissions would be rerun for final evaluation, and that these reruns could lead to outcomes different from what participants inferred from the visible public notebook scores. That part seems to rely more on Kaggle’s notebook competition mechanics and forum clarifications than on a clearly stated rule.", "votes": 3, "replies": [ { "id": 3432825, "postDate": "2026-04-01T00:03:28.797Z", "content": "you are right,l feel disappointed really
", "rawMarkdown": "you are right,l feel disappointed really", "votes": 2 } ] }, { "id": 3432621, "postDate": "2026-03-31T20:09:26.857Z", "content": " Am I missing something? We selected this submission, but on the leaderboard same notebook is showing with 2 different scores a 1 is lower score and 1 is higher. Like were there multiple reruns of the same notebook
yeah, same with me my selected notebook wasn't selected for some reason.
", "rawMarkdown": "yeah, same with me my selected notebook wasn't selected for some reason." }, { "id": 3432627, "postDate": "2026-03-31T20:12:45.563Z", "rawMarkdown": "", "votes": -1, "isDeleted": true, "replies": [ { "id": 3432657, "postDate": "2026-03-31T20:34:40.503Z", "content": "But we got kicked out before the rerun.
", "rawMarkdown": "But we got kicked out before the rerun.", "votes": 1 }, { "id": 3432661, "postDate": "2026-03-31T20:39:07.850Z", "rawMarkdown": "", "isDeleted": true } ] }, { "id": 3432632, "postDate": "2026-03-31T20:17:12.680Z", "content": "The notebooks you selected were rerun. The name should be the same but the description will say \"notebook rerun\"
\nAlthough, on this topic, my private scores before the rerun were much higher. Not totally sure what that is about
", "rawMarkdown": "The notebooks you selected were rerun. The name should be the same but the description will say \"notebook rerun\"\n\nAlthough, on this topic, my private scores before the rerun were much higher. Not totally sure what that is about", "replies": [ { "id": 3432682, "postDate": "2026-03-31T20:53:23.520Z", "content": "Thanks for the response @jackvd.
\nFor the difference in private scores after the rerun, please check out this thread: https://www.kaggle.com/competitions/stanford-rna-3d-folding-2/discussion/686601
", "rawMarkdown": "Thanks for the response @jackvd.\n\nFor the difference in private scores after the rerun, please check out this thread: https://www.kaggle.com/competitions/stanford-rna-3d-folding-2/discussion/686601", "votes": -1 }, { "id": 3432691, "postDate": "2026-03-31T21:06:01.247Z", "content": "Yes, my comment was before your post - thanks for clarifying!
", "rawMarkdown": "Yes, my comment was before your post - thanks for clarifying!" } ] } ] }, { "id": 3432956, "postDate": "2026-04-01T03:33:10.577Z", "content": "Dear Hosts,
\nI would like to further evaluate and validate some approaches that I did not ultimately select. Could you please confirm whether late submissions will not be available until the latter half of 2026?
\nI would appreciate your response at your convenience.
", "rawMarkdown": "Dear Hosts,\n\nI would like to further evaluate and validate some approaches that I did not ultimately select. Could you please confirm whether late submissions will not be available until the latter half of 2026?\n\nI would appreciate your response at your convenience.\n", "votes": 1, "replies": [ { "id": 3434102, "postDate": "2026-04-02T15:42:17.137Z", "rawMarkdown": "", "isDeleted": true }, { "id": 3434168, "postDate": "2026-04-02T17:06:59.310Z", "content": "The timeline for reopening the competition will depend on some decisions from experimental collaborators. We expect it to happen in 2026 but we can't make any promises on how soon. Please stay tuned to this Discussion.
", "rawMarkdown": "The timeline for reopening the competition will depend on some decisions from experimental collaborators. We expect it to happen in 2026 but we can't make any promises on how soon. Please stay tuned to this Discussion.", "votes": 2 } ] }, { "id": 3432709, "postDate": "2026-03-31T21:33:55.853Z", "content": "To the organizers:
\nThank you for hosting this competition.
\nI participated in both Part 1 and Part 2, and personally, I feel that I was able to approach the task from a different perspective in Part 2. I think that was true throughout the competition as a whole. It was a very stimulating three months.
\nI'm looking forward to Part 3!
", "rawMarkdown": "To the organizers:\n\nThank you for hosting this competition.\n\nI participated in both Part 1 and Part 2, and personally, I feel that I was able to approach the task from a different perspective in Part 2. I think that was true throughout the competition as a whole. It was a very stimulating three months.\n\nI'm looking forward to Part 3!" }, { "id": 3432688, "postDate": "2026-03-31T21:02:06.197Z", "content": "Dear hosts,\nWhile I struggled to find new solutions for the more challenging structures, this experience has been an invaluable learning opportunity. Thank you very much for your support and guidance.
", "rawMarkdown": "Dear hosts,\nWhile I struggled to find new solutions for the more challenging structures, this experience has been an invaluable learning opportunity. Thank you very much for your support and guidance." }, { "id": 3432979, "postDate": "2026-04-01T04:12:47.017Z", "content": "My goal was to never worry about the public score board. Overfitting to that, I realized early on, was futile and not a strategy !!
", "rawMarkdown": "My goal was to never worry about the public score board. Overfitting to that, I realized early on, was futile and not a strategy !! ", "votes": 1, "replies": [ { "id": 3433145, "postDate": "2026-04-01T09:34:58.403Z", "content": "Me too.
\nI had bet they came from high-scoring public notebook optimized for public leaderboard.
Dear @rhijudas sir the actual problem is that it didn't selected my top two notebooks . That is my concern . You can check this screenshot![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F29100267%2F646d60d0cbacd5b0509f950afbc83440%2Finbox_29100267_7d1da71cb554dd99d2d85af25fefea91_Stanford_LB.png?generation=1776279866492204&alt=media\")
Hi, the top 2 public LB notebooks were correctly selected, but the Public scores changed slightly on the final 03/31/2026 rerun because the notebooks did not run deterministically.
", "rawMarkdown": "Hi, the top 2 public LB notebooks were correctly selected, but the Public scores changed slightly on the final 03/31/2026 rerun because the notebooks did not run deterministically. ", "replies": [ { "id": 3442991, "postDate": "2026-04-16T02:54:58.513Z", "content": "@rhijudas Thank you so much for clearing my doubts regarding the scoring. \nAs my code has still beaten the best_template_oracle I am giving access to my notebook if this could be of any help to you . https://www.kaggle.com/code/techtech69/stanford-rnafolding?scriptVersionId=305754975
\nThank you for hosting such a great competition.
", "rawMarkdown": "@rhijudas Thank you so much for clearing my doubts regarding the scoring. \nAs my code has still beaten the best_template_oracle I am giving access to my notebook if this could be of any help to you . https://www.kaggle.com/code/techtech69/stanford-rnafolding?scriptVersionId=305754975\n\nThank you for hosting such a great competition. " } ] } ] }, { "id": 3433732, "postDate": "2026-04-02T03:38:20.823Z", "content": "Dear Hosts ,\nI would like to bring you attention to this as kaggle says that it auto selects our best public scoring notebooks as final selection. But why wasn't my best performing notebook wasnt selected? Even when it had a score greater than 0.41996 in public run ?
", "rawMarkdown": "Dear Hosts ,\nI would like to bring you attention to this as kaggle says that it auto selects our best public scoring notebooks as final selection. But why wasn't my best performing notebook wasnt selected? Even when it had a score greater than 0.41996 in public run ?", "replies": [ { "id": 3433736, "postDate": "2026-04-02T03:47:05.213Z", "content": "@rhijudas @youhanlee @przemekporebski @theoviel @walterreade @chrismunley @ashleyoldacre .
", "rawMarkdown": "@rhijudas @youhanlee @przemekporebski @theoviel @walterreade @chrismunley @ashleyoldacre ." }, { "id": 3439860, "postDate": "2026-04-11T10:56:40.470Z", "content": "Hello organisers , no one has responded to my query yet ? @rhijudas
", "rawMarkdown": "Hello organisers , no one has responded to my query yet ? @rhijudas ", "replies": [ { "id": 3439928, "postDate": "2026-04-11T14:37:35.457Z", "content": "Hi, your top two public score notebooks were rerun. The public scores shifted slightly due to nondeterministic behavior in the notebooks.
", "rawMarkdown": "Hi, your top two public score notebooks were rerun. The public scores shifted slightly due to nondeterministic behavior in the notebooks." }, { "id": 3440587, "postDate": "2026-04-12T17:09:57.340Z", "content": "Dear @rhijudas sir the actual problem is that it didn't selected my top two notebooks . That is my concern . You can check this screenshot![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F29100267%2F7d1da71cb554dd99d2d85af25fefea91%2FStanford_LB.png?generation=1776013786237481&alt=media\")
Hello @rhijudas sir can you please check this issue.
", "rawMarkdown": "Hello @rhijudas sir can you please check this issue. " }, { "id": 3442660, "postDate": "2026-04-15T18:45:21.933Z", "content": "Hello @rhijudas sir can you please check this out . I have put a lot of work into this
", "rawMarkdown": "Hello @rhijudas sir can you please check this out . I have put a lot of work into this" } ] } ] }, { "id": 3433194, "postDate": "2026-04-01T11:28:51.003Z", "content": "Dear @rhijudas. This was my first Kaggle competition and I had no idea that I MUST select the 2 submissions to be evaluated for the final leaderboard score, as Kaggle explicitly mentioned that the top two public scoring submissions would be autoselected for submission. For the final leaderboard submission, Kaggle selected two random low (public) scoring submissions for the leaderboard. Is there a chance for my two top public scoring notebooks be reconsidered for the leaderboard? Thanks!
", "rawMarkdown": "Dear @rhijudas. This was my first Kaggle competition and I had no idea that I MUST select the 2 submissions to be evaluated for the final leaderboard score, as Kaggle explicitly mentioned that the top two public scoring submissions would be autoselected for submission. For the final leaderboard submission, Kaggle selected two random low (public) scoring submissions for the leaderboard. Is there a chance for my two top public scoring notebooks be reconsidered for the leaderboard? Thanks!" }, { "id": 3433173, "postDate": "2026-04-01T10:37:54.290Z", "content": "Congratulations to the winners, if only anyone taught me how to win. Antigravity did not do anything for me. How long does it take to learn this data science?
", "rawMarkdown": "Congratulations to the winners, if only anyone taught me how to win. Antigravity did not do anything for me. How long does it take to learn this data science?" }, { "id": 3433069, "postDate": "2026-04-01T06:29:46.197Z", "content": "Hi, professor. Thank you for hosting this competition in this year.
\nI have a question. Shouldn't the CASP evaluation metrics include multiple indicators, including TM-score, and unlike the Kaggle competition, CASP requires the score for each individual coordinate, not the best score among the five coordinates? The gold medalist's solution hasn't been disclosed yet, and I see that most solutions currently use SOTA ensemble strategies. Essentially, they might not know which coordinate yields the best score. Additionally, the public and private leaderboard scores fluctuate quite a bit. So, how are you planning to use the competition results to prepare for the CASP competition? Therefore, I'm also very curious to see if the gold medalist's solution can clarify my doubts. I personally think this is a noteworthy issue in this competition.
", "rawMarkdown": "Hi, professor. Thank you for hosting this competition in this year.\n\nI have a question. Shouldn't the CASP evaluation metrics include multiple indicators, including TM-score, and unlike the Kaggle competition, CASP requires the score for each individual coordinate, not the best score among the five coordinates? The gold medalist's solution hasn't been disclosed yet, and I see that most solutions currently use SOTA ensemble strategies. Essentially, they might not know which coordinate yields the best score. Additionally, the public and private leaderboard scores fluctuate quite a bit. So, how are you planning to use the competition results to prepare for the CASP competition? Therefore, I'm also very curious to see if the gold medalist's solution can clarify my doubts. I personally think this is a noteworthy issue in this competition." }, { "id": 3432996, "postDate": "2026-04-01T04:44:51.433Z", "content": "On the consistency between public and private leaderboards
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F25984450%2F8f3964808968cade9341bf0036a1e877%2F23333ScreenShot_2026-04-01_120954_888.png?generation=1775018688885906&alt=media\")
and this\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F25984450%2F40e4c7bb21f6a044e9fe7c1e3652dce4%2F06_public_vs_private_scatter_trend.png?generation=1775019310429051&alt=media\")
Will there be a Third Competition 😼😼😼
", "rawMarkdown": "Will there be a Third Competition 😼😼😼" }, { "id": 3432689, "postDate": "2026-03-31T21:02:46.110Z", "content": "![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F13686656%2F7ea92a17041870aa00dddf72c9e15624%2FScreenshot_1-4-2026_35951_www.kaggle.com.jpeg?generation=1774990891533073&alt=media\")
i am sorry mr i have a question my chosing selection notebook before deadline in private score is 520 but in final private leaderboard is change,thank you
", "rawMarkdown": "\n\ni am sorry mr i have a question my chosing selection notebook before deadline in private score is 520 but in final private leaderboard is change,thank you", "replies": [ { "id": 3432714, "postDate": "2026-03-31T21:42:31.993Z", "content": "For the difference in private scores after the rerun, please check out this thread: https://www.kaggle.com/competitions/stanford-rna-3d-folding-2/discussion/686601
", "rawMarkdown": "For the difference in private scores after the rerun, please check out this thread: https://www.kaggle.com/competitions/stanford-rna-3d-folding-2/discussion/686601" } ] }, { "id": 3432672, "postDate": "2026-03-31T20:46:07.933Z", "content": "Can you please help me understand if this notebook has a final private score, or if it wasn't rerun?\nhttps://www.kaggle.com/code/analyticaobscura/rna-3d-folding-tbm
\nI still feel a misunderstanding of what our team violated. This notebook was opened for the public only after the submission deadline, and support only responded with the following, which makes me think this could be just an automatic response with no human verification
\n\n", "rawMarkdown": "Can you please help me understand if this notebook has a final private score, or if it wasn't rerun?\nhttps://www.kaggle.com/code/analyticaobscura/rna-3d-folding-tbm\n\nI still feel a misunderstanding of what our team violated. This notebook was opened for the public only after the submission deadline, and support only responded with the following, which makes me think this could be just an automatic response with no human verification\n> Thank you for your inquiry, Your team was removed for breaking the competition rules. Unfortunately we can not disclose which rules were broken, but you are welcome to check with your team to inquire as to why the removal was necessary.", "replies": [ { "id": 3432681, "postDate": "2026-03-31T20:51:37.377Z", "content": "Thank you for your inquiry, Your team was removed for breaking the competition rules. Unfortunately we can not disclose which rules were broken, but you are welcome to check with your team to inquire as to why the removal was necessary.
\n
@honganzhu look, their private score is also above 0.503.
", "rawMarkdown": "@honganzhu look, their private score is also above 0.503.", "replies": [ { "id": 3432683, "postDate": "2026-03-31T20:55:38.517Z", "rawMarkdown": "", "votes": 1, "isDeleted": true }, { "id": 3432823, "postDate": "2026-04-01T00:02:29.483Z", "content": "me too,mine is 0.540
", "rawMarkdown": "me too,mine is 0.540" } ] }, { "id": 3432687, "postDate": "2026-03-31T21:01:16.050Z", "content": "@rhijudas , can you please check this notebook and the hypothesis below? If so, is there any Kaggle verification for that instead of automatic disqualification?
\n\n", "rawMarkdown": "@rhijudas , can you please check this notebook and the hypothesis below? If so, is there any Kaggle verification for that instead of automatic disqualification?\n\n> that's what i was saying. a high score in the \"small test set before rerun\" will get you disqualified. On the other hand, for the normal teams, if their notebook is random, they just ran it multiple times till the final private score is below the benchmark, this is very evident from the first comment above", "replies": [ { "id": 3433760, "postDate": "2026-04-02T04:52:45.743Z", "content": "that's what i was saying. a high score in the \"small test set before rerun\" will get you disqualified. On the other hand, for the normal teams, if their notebook is random, they just ran it multiple times till the final private score is below the benchmark, this is very evident from the first comment above
\n
All my submission is warning but it could run as I participating, WHY? \n![\"socre\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F32698250%2Fc47e7997dc521d10ef58789c86b8ef3d%2F2026-04-02%2012.51.46.png?generation=1775105541012163&alt=media\")
Congratulations on your method's performance.
\nYou are very welcome to write an independent research paper on the methodology and results. Several prior top Kagglers have done so, with outstanding impact. As an example, @shujun717 wrote a classic paper on mRNA degradation and also contributed to the collective paper on that Kaggle challenge.
\nWe look forward to hearing more about your solution!
", "rawMarkdown": "Congratulations on your method's performance. \n\nYou are very welcome to write an independent research paper on the methodology and results. Several prior top Kagglers have done so, with outstanding impact. As an example, @shujun717 wrote a classic paper on [mRNA degradation](https://pubmed.ncbi.nlm.nih.gov/36633966/) and also contributed to the [collective paper](https://www.nature.com/articles/s42256-022-00571-8) on that Kaggle challenge. \n\nWe look forward to hearing more about your solution!", "replies": [ { "id": 3432897, "postDate": "2026-04-01T02:05:10.537Z", "content": "ok thank you
", "rawMarkdown": " ok thank you" } ] } ] }, { "id": 3432880, "postDate": "2026-04-01T01:43:41.270Z", "rawMarkdown": "", "isDeleted": true }, { "id": 3432879, "postDate": "2026-04-01T01:42:35.217Z", "rawMarkdown": "", "isDeleted": true }, { "id": 3432822, "postDate": "2026-04-01T00:01:35.737Z", "rawMarkdown": "", "isDeleted": true }, { "id": 3432656, "postDate": "2026-03-31T20:34:36.907Z", "rawMarkdown": "", "isDeleted": true, "replies": [ { "id": 3432658, "postDate": "2026-03-31T20:36:06.940Z", "rawMarkdown": "", "isDeleted": true } ] }, { "id": 3432603, "postDate": "2026-03-31T19:56:48.393Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 3432700, "author_name": "ömer kiraz", "author_url": "", "post_date": "2026-03-31T21:21:21.403000", "content": "I understand that auto-selection and final Private Leaderboard evaluation are supported by the rules. However, I do not think the rules made it sufficiently explicit that notebook submissions would be rerun for final evaluation, and that these reruns could lead to outcomes different from what participants inferred from the visible public notebook scores. That part seems to rely more on Kaggle’s notebook competition mechanics and forum clarifications than on a clearly stated rule.
", "votes": 3, "replies": [ { "id": 3432825, "author_name": "grifth", "author_url": "", "post_date": "2026-04-01T00:03:28.797000", "content": "you are right,l feel disappointed really
", "votes": 2, "replies": [] } ] }, { "id": 3432621, "author_name": "", "author_url": "", "post_date": "2026-03-31T20:09:26.857000", "content": " Am I missing something? We selected this submission, but on the leaderboard same notebook is showing with 2 different scores a 1 is lower score and 1 is higher. Like were there multiple reruns of the same notebook
yeah, same with me my selected notebook wasn't selected for some reason.
", "votes": 0, "replies": [] }, { "id": 3432627, "author_name": "", "author_url": "", "post_date": "2026-03-31T20:12:45.563000", "content": "", "votes": -1, "replies": [ { "id": 3432657, "author_name": "Andrey Zhabsky", "author_url": "", "post_date": "2026-03-31T20:34:40.503000", "content": "But we got kicked out before the rerun.
", "votes": 1, "replies": [] }, { "id": 3432661, "author_name": "", "author_url": "", "post_date": "2026-03-31T20:39:07.850000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 3432632, "author_name": "Jack", "author_url": "", "post_date": "2026-03-31T20:17:12.680000", "content": "The notebooks you selected were rerun. The name should be the same but the description will say \"notebook rerun\"
\nAlthough, on this topic, my private scores before the rerun were much higher. Not totally sure what that is about
", "votes": 0, "replies": [ { "id": 3432682, "author_name": "Rhiju Das", "author_url": "", "post_date": "2026-03-31T20:53:23.520000", "content": "Thanks for the response @jackvd.
\nFor the difference in private scores after the rerun, please check out this thread: https://www.kaggle.com/competitions/stanford-rna-3d-folding-2/discussion/686601
", "votes": -1, "replies": [] }, { "id": 3432691, "author_name": "Jack", "author_url": "", "post_date": "2026-03-31T21:06:01.247000", "content": "Yes, my comment was before your post - thanks for clarifying!
", "votes": 0, "replies": [] } ] } ] }, { "id": 3432956, "author_name": "Shun Kuraishi", "author_url": "", "post_date": "2026-04-01T03:33:10.577000", "content": "Dear Hosts,
\nI would like to further evaluate and validate some approaches that I did not ultimately select. Could you please confirm whether late submissions will not be available until the latter half of 2026?
\nI would appreciate your response at your convenience.
", "votes": 1, "replies": [ { "id": 3434102, "author_name": "", "author_url": "", "post_date": "2026-04-02T15:42:17.137000", "content": "", "votes": 0, "replies": [] }, { "id": 3434168, "author_name": "Rhiju Das", "author_url": "", "post_date": "2026-04-02T17:06:59.310000", "content": "The timeline for reopening the competition will depend on some decisions from experimental collaborators. We expect it to happen in 2026 but we can't make any promises on how soon. Please stay tuned to this Discussion.
", "votes": 2, "replies": [] } ] }, { "id": 3432709, "author_name": "koooeo", "author_url": "", "post_date": "2026-03-31T21:33:55.853000", "content": "To the organizers:
\nThank you for hosting this competition.
\nI participated in both Part 1 and Part 2, and personally, I feel that I was able to approach the task from a different perspective in Part 2. I think that was true throughout the competition as a whole. It was a very stimulating three months.
\nI'm looking forward to Part 3!
", "votes": 0, "replies": [] }, { "id": 3432688, "author_name": "Shun Kuraishi", "author_url": "", "post_date": "2026-03-31T21:02:06.197000", "content": "Dear hosts,\nWhile I struggled to find new solutions for the more challenging structures, this experience has been an invaluable learning opportunity. Thank you very much for your support and guidance.
", "votes": 0, "replies": [] }, { "id": 3432979, "author_name": "Alan T. Andrea", "author_url": "", "post_date": "2026-04-01T04:12:47.017000", "content": "My goal was to never worry about the public score board. Overfitting to that, I realized early on, was futile and not a strategy !!
", "votes": 1, "replies": [ { "id": 3433145, "author_name": "odat", "author_url": "", "post_date": "2026-04-01T09:34:58.403000", "content": "Me too.
\nI had bet they came from high-scoring public notebook optimized for public leaderboard.
Dear @rhijudas sir the actual problem is that it didn't selected my top two notebooks . That is my concern . You can check this screenshot
Hi, the top 2 public LB notebooks were correctly selected, but the Public scores changed slightly on the final 03/31/2026 rerun because the notebooks did not run deterministically.
", "votes": 0, "replies": [ { "id": 3442991, "author_name": "techtech69", "author_url": "", "post_date": "2026-04-16T02:54:58.513000", "content": "@rhijudas Thank you so much for clearing my doubts regarding the scoring. \nAs my code has still beaten the best_template_oracle I am giving access to my notebook if this could be of any help to you . https://www.kaggle.com/code/techtech69/stanford-rnafolding?scriptVersionId=305754975
\nThank you for hosting such a great competition.
", "votes": 0, "replies": [] } ] } ] }, { "id": 3433732, "author_name": "techtech69", "author_url": "", "post_date": "2026-04-02T03:38:20.823000", "content": "Dear Hosts ,\nI would like to bring you attention to this as kaggle says that it auto selects our best public scoring notebooks as final selection. But why wasn't my best performing notebook wasnt selected? Even when it had a score greater than 0.41996 in public run ?
", "votes": 0, "replies": [ { "id": 3433736, "author_name": "techtech69", "author_url": "", "post_date": "2026-04-02T03:47:05.213000", "content": "@rhijudas @youhanlee @przemekporebski @theoviel @walterreade @chrismunley @ashleyoldacre .
", "votes": 0, "replies": [] }, { "id": 3439860, "author_name": "techtech69", "author_url": "", "post_date": "2026-04-11T10:56:40.470000", "content": "Hello organisers , no one has responded to my query yet ? @rhijudas
", "votes": 0, "replies": [ { "id": 3439928, "author_name": "Rhiju Das", "author_url": "", "post_date": "2026-04-11T14:37:35.457000", "content": "Hi, your top two public score notebooks were rerun. The public scores shifted slightly due to nondeterministic behavior in the notebooks.
", "votes": 0, "replies": [] }, { "id": 3440587, "author_name": "techtech69", "author_url": "", "post_date": "2026-04-12T17:09:57.340000", "content": "Dear @rhijudas sir the actual problem is that it didn't selected my top two notebooks . That is my concern . You can check this screenshot
Hello @rhijudas sir can you please check this issue.
", "votes": 0, "replies": [] }, { "id": 3442660, "author_name": "techtech69", "author_url": "", "post_date": "2026-04-15T18:45:21.933000", "content": "Hello @rhijudas sir can you please check this out . I have put a lot of work into this
", "votes": 0, "replies": [] } ] } ] }, { "id": 3433194, "author_name": "epigene", "author_url": "", "post_date": "2026-04-01T11:28:51.003000", "content": "Dear @rhijudas. This was my first Kaggle competition and I had no idea that I MUST select the 2 submissions to be evaluated for the final leaderboard score, as Kaggle explicitly mentioned that the top two public scoring submissions would be autoselected for submission. For the final leaderboard submission, Kaggle selected two random low (public) scoring submissions for the leaderboard. Is there a chance for my two top public scoring notebooks be reconsidered for the leaderboard? Thanks!
", "votes": 0, "replies": [] }, { "id": 3433173, "author_name": "Assistant to JJ", "author_url": "", "post_date": "2026-04-01T10:37:54.290000", "content": "Congratulations to the winners, if only anyone taught me how to win. Antigravity did not do anything for me. How long does it take to learn this data science?
", "votes": 0, "replies": [] }, { "id": 3433069, "author_name": "Timmy Juicehouse", "author_url": "", "post_date": "2026-04-01T06:29:46.197000", "content": "Hi, professor. Thank you for hosting this competition in this year.
\nI have a question. Shouldn't the CASP evaluation metrics include multiple indicators, including TM-score, and unlike the Kaggle competition, CASP requires the score for each individual coordinate, not the best score among the five coordinates? The gold medalist's solution hasn't been disclosed yet, and I see that most solutions currently use SOTA ensemble strategies. Essentially, they might not know which coordinate yields the best score. Additionally, the public and private leaderboard scores fluctuate quite a bit. So, how are you planning to use the competition results to prepare for the CASP competition? Therefore, I'm also very curious to see if the gold medalist's solution can clarify my doubts. I personally think this is a noteworthy issue in this competition.
", "votes": 0, "replies": [] }, { "id": 3432996, "author_name": "Aleck", "author_url": "", "post_date": "2026-04-01T04:44:51.433000", "content": "On the consistency between public and private leaderboards
\n
and this\n
Will there be a Third Competition 😼😼😼
", "votes": 0, "replies": [] }, { "id": 3432689, "author_name": "Hellboy", "author_url": "", "post_date": "2026-03-31T21:02:46.110000", "content": "
i am sorry mr i have a question my chosing selection notebook before deadline in private score is 520 but in final private leaderboard is change,thank you
", "votes": 0, "replies": [ { "id": 3432714, "author_name": "Rhiju Das", "author_url": "", "post_date": "2026-03-31T21:42:31.993000", "content": "For the difference in private scores after the rerun, please check out this thread: https://www.kaggle.com/competitions/stanford-rna-3d-folding-2/discussion/686601
", "votes": 0, "replies": [] } ] }, { "id": 3432672, "author_name": "apat1n", "author_url": "", "post_date": "2026-03-31T20:46:07.933000", "content": "Can you please help me understand if this notebook has a final private score, or if it wasn't rerun?\nhttps://www.kaggle.com/code/analyticaobscura/rna-3d-folding-tbm
\nI still feel a misunderstanding of what our team violated. This notebook was opened for the public only after the submission deadline, and support only responded with the following, which makes me think this could be just an automatic response with no human verification
\n\n", "votes": 0, "replies": [ { "id": 3432681, "author_name": "Andrey Zhabsky", "author_url": "", "post_date": "2026-03-31T20:51:37.377000", "content": "Thank you for your inquiry, Your team was removed for breaking the competition rules. Unfortunately we can not disclose which rules were broken, but you are welcome to check with your team to inquire as to why the removal was necessary.
\n
@honganzhu look, their private score is also above 0.503.
", "votes": 0, "replies": [ { "id": 3432683, "author_name": "", "author_url": "", "post_date": "2026-03-31T20:55:38.517000", "content": "", "votes": 1, "replies": [] }, { "id": 3432823, "author_name": "grifth", "author_url": "", "post_date": "2026-04-01T00:02:29.483000", "content": "me too,mine is 0.540
", "votes": 0, "replies": [] } ] }, { "id": 3432687, "author_name": "apat1n", "author_url": "", "post_date": "2026-03-31T21:01:16.050000", "content": "@rhijudas , can you please check this notebook and the hypothesis below? If so, is there any Kaggle verification for that instead of automatic disqualification?
\n\n", "votes": 0, "replies": [ { "id": 3433760, "author_name": "星乃一歌", "author_url": "", "post_date": "2026-04-02T04:52:45.743000", "content": "that's what i was saying. a high score in the \"small test set before rerun\" will get you disqualified. On the other hand, for the normal teams, if their notebook is random, they just ran it multiple times till the final private score is below the benchmark, this is very evident from the first comment above
\n
All my submission is warning but it could run as I participating, WHY? \n
Congratulations on your method's performance.
\nYou are very welcome to write an independent research paper on the methodology and results. Several prior top Kagglers have done so, with outstanding impact. As an example, @shujun717 wrote a classic paper on mRNA degradation and also contributed to the collective paper on that Kaggle challenge.
\nWe look forward to hearing more about your solution!
", "votes": 0, "replies": [ { "id": 3432897, "author_name": "huyang111", "author_url": "", "post_date": "2026-04-01T02:05:10.537000", "content": "ok thank you
", "votes": 0, "replies": [] } ] } ] }, { "id": 3432880, "author_name": "", "author_url": "", "post_date": "2026-04-01T01:43:41.270000", "content": "", "votes": 0, "replies": [] }, { "id": 3432879, "author_name": "", "author_url": "", "post_date": "2026-04-01T01:42:35.217000", "content": "", "votes": 0, "replies": [] }, { "id": 3432822, "author_name": "", "author_url": "", "post_date": "2026-04-01T00:01:35.737000", "content": "", "votes": 0, "replies": [] }, { "id": 3432656, "author_name": "", "author_url": "", "post_date": "2026-03-31T20:34:36.907000", "content": "", "votes": 0, "replies": [ { "id": 3432658, "author_name": "", "author_url": "", "post_date": "2026-03-31T20:36:06.940000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 3432603, "author_name": "", "author_url": "", "post_date": "2026-03-31T19:56:48.393000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "3432600": "After rerun of team’s two selected notebooks on a completely hidden test set, the [final Private Leaderboard](https://www.kaggle.com/competitions/stanford-rna-3d-folding-2/leaderboard) has posted!\n\n**The leaderboard**\n\nWe are pleased to announce that **team_cp** @naganohikaru @yutaroito, **AyPy** @AyPy, and **Stefan Stefanov** @stefanstefanov are our three Leaderboard Prize winners. It was a close finish — congratulations to all who submitted! \n\nWe were impressed that this competition elicited particularly strong [publicly shared notebooks](https://www.kaggle.com/competitions/stanford-rna-3d-folding-2/code) that many folks could build on and take advantage of. \n\nSo it has been particularly exciting to now see several teams who were lower in the Public Leaderboard come out near the top of the Private Leaderboard — we are very curious to hear their strategies for generalization. \n\n\n**Competition goals: Complex targets and beating the `best_template_oracle`**\n\nThere were several goals of this Part 2 competition.\n\nFirst, we wanted to test if automated modeling methods could generalize to RNAs without previously available templates and to other complex RNA targets, both aspects missing from [Part 1](https://www.kaggle.com/competitions/stanford-rna-3d-folding/overview). \n\nBased on the results, the answer appears to be a resounding yes! We look forward to presenting detailed 3D structural comparisons later in 2026, when the ground truth structures will be made public.\n\nSecond, we were curious if Kagglers could advance RNA structure prediction to, for the first time, consistently outperform a very hard baseline: the `best_template_oracle`. \n\nThis baseline was assembled from the best publicly available prior 3D template for each target – with special knowledge of the ground truth structure. In the Part 1 competition, Kaggle notebooks could *not* beat the best template oracle, even when all Kaggle structure predictions were pooled (see Part 1 [preprint](https://www.biorxiv.org/content/10.64898/2025.12.30.696949v1.full)).\n\nHere are the results from Part 2:\n\n- 0.562 - Top 20 notebooks\n- 0.515 - Top 2 (`team_cp` + `AyPy`)\n- 0.503 - `best_template_oracle`\n- 0.497 - `team_cp`\n- 0.492 - `AyPy`\n\nThe top two scores from individual `team_cp` and `AyPy` notebooks are very close to the best template oracle — they are almost within statistical fluctuations (about 0.006). \n\nNow look at the `Top 2` model — it simply pools the predictions of one notebook each from `team_cp` and `AyPy`. Excitingly, this Top 2 outperforms the best template oracle! 🎉\n\nEven better, if we take the Top 20 submitted notebooks, their collective best prediction for each target very convincingly outperforms the template oracle. Different top groups did better on different targets, demonstrating yet again the power of the Kaggle community to find diverse, complementary solutions to hard scientific problems.\n\n**What’s next?**\n\nIn just a couple weeks, the world’s flagship competition for molecular modeling will begin: the 17th Critical Assessment for Structure Prediction ([CASP17](https://predictioncenter.org/casp17/index.cgi)).\n\nCould this year mark a special jump in the accuracy of automated RNA structure prediction? The rapid advances in the last two Kaggle challenges certainly give us optimism — and your efforts in the next two weeks could make the difference:\n\n*This is the time to make your notebooks public and to write up your solutions!*\n\nWe’re particularly interested in teams who placed in the Top 20 because, as noted above, the solutions appear complementary. \n\nIf we share our insights and codes, we should be able to make an RNA structure prediction model that is better than the sum of its parts and increase our impact across biology, medicine, and biotechnology. \n\nWe’re excited to again have collaboration across Kagglers, Stanford, HHMI, and NVIDIA for this ‘model synthesis’ effort and to enter resulting models into CASP17 for further rigorous blind tests. Looking forward to seeing you on the forums, in CASP, and in the RNA world!\n \nFor the hosts,\n\nRhiju Das @rhijudas and Przemek Porebski @przemekporebski\n\nAnd, again, many thanks to our co-hosts and collaborators:\n* The AI@HHMI initiative at Janelia Research Campus, the Toor lab at U.C.San Diego, and additional RNA cryo-EM groups providing blind targets.\n* NVIDIA’s BioNemo and KGMON, especially @youhanlee @chrismunley @tviel\n* Kaggle devs: Walter Reade @inversion and Ashley Oldacre @ashleyoldacre\n\n", "3432700": "I understand that auto-selection and final Private Leaderboard evaluation are supported by the rules. However, I do not think the rules made it sufficiently explicit that notebook submissions would be rerun for final evaluation, and that these reruns could lead to outcomes different from what participants inferred from the visible public notebook scores. That part seems to rely more on Kaggle’s notebook competition mechanics and forum clarifications than on a clearly stated rule.", "3432621": " Am I missing something? We selected this submission, but on the leaderboard same notebook is showing with 2 different scores a 1 is lower score and 1 is higher. Like were there multiple reruns of the same notebook ", "3432956": "Dear Hosts,\n\nI would like to further evaluate and validate some approaches that I did not ultimately select. Could you please confirm whether late submissions will not be available until the latter half of 2026?\n\nI would appreciate your response at your convenience.\n", "3432709": "To the organizers:\n\nThank you for hosting this competition.\n\nI participated in both Part 1 and Part 2, and personally, I feel that I was able to approach the task from a different perspective in Part 2. I think that was true throughout the competition as a whole. It was a very stimulating three months.\n\nI'm looking forward to Part 3!", "3432688": "Dear hosts,\nWhile I struggled to find new solutions for the more challenging structures, this experience has been an invaluable learning opportunity. Thank you very much for your support and guidance.", "3432979": "My goal was to never worry about the public score board. Overfitting to that, I realized early on, was futile and not a strategy !! ", "3442685": "Dear @rhijudas sir the actual problem is that it didn't selected my top two notebooks . That is my concern . You can check this screenshot", "3433732": "Dear Hosts ,\nI would like to bring you attention to this as kaggle says that it auto selects our best public scoring notebooks as final selection. But why wasn't my best performing notebook wasnt selected? Even when it had a score greater than 0.41996 in public run ?", "3433194": "Dear @rhijudas. This was my first Kaggle competition and I had no idea that I MUST select the 2 submissions to be evaluated for the final leaderboard score, as Kaggle explicitly mentioned that the top two public scoring submissions would be autoselected for submission. For the final leaderboard submission, Kaggle selected two random low (public) scoring submissions for the leaderboard. Is there a chance for my two top public scoring notebooks be reconsidered for the leaderboard? Thanks!", "3433173": "Congratulations to the winners, if only anyone taught me how to win. Antigravity did not do anything for me. How long does it take to learn this data science?", "3433069": "Hi, professor. Thank you for hosting this competition in this year.\n\nI have a question. Shouldn't the CASP evaluation metrics include multiple indicators, including TM-score, and unlike the Kaggle competition, CASP requires the score for each individual coordinate, not the best score among the five coordinates? The gold medalist's solution hasn't been disclosed yet, and I see that most solutions currently use SOTA ensemble strategies. Essentially, they might not know which coordinate yields the best score. Additionally, the public and private leaderboard scores fluctuate quite a bit. So, how are you planning to use the competition results to prepare for the CASP competition? Therefore, I'm also very curious to see if the gold medalist's solution can clarify my doubts. I personally think this is a noteworthy issue in this competition.", "3432996": "On the consistency between public and private leaderboards\n\n", "3432787": "Will there be a Third Competition 😼😼😼", "3432689": "\n\ni am sorry mr i have a question my chosing selection notebook before deadline in private score is 520 but in final private leaderboard is change,thank you", "3432672": "Can you please help me understand if this notebook has a final private score, or if it wasn't rerun?\nhttps://www.kaggle.com/code/analyticaobscura/rna-3d-folding-tbm\n\nI still feel a misunderstanding of what our team violated. This notebook was opened for the public only after the submission deadline, and support only responded with the following, which makes me think this could be just an automatic response with no human verification\n> Thank you for your inquiry, Your team was removed for breaking the competition rules. Unfortunately we can not disclose which rules were broken, but you are welcome to check with your team to inquire as to why the removal was necessary.", "3478661": "", "3442684": "", "3432947": "", "3432881": "", "3432880": "", "3432879": "", "3432822": "", "3432656": "", "3432603": "" } }